Conventional forward error correction (FEC) code rate decision schemes using analytical source coding distortion model and channel-induced distortion model are usually complex, and require the typical process of model parameter training which involves potentially high computational complexity and implementation cost. To avoid the complex modeling procedure, we propose a simple but accurate joint source-channel distortion model to estimate channel loss threshold set for optimal FEC code rate decision.
When transporting video over the wireline/wireless channel, each video packet can be received correctly, received with errors, or lost by the network. It is necessary for the video sender to provide adequate error resilience features. It is efficient to adaptively change the FEC code rate in response to time-varying packet loss dynamics in order to ensure consistent optimal video quality
. The optimal FEC code rate decision is a crucial procedure to determine the optimal source and channel coding rates to minimize the overall picture distortion when transporting video packets over packet loss channels. Several previous studies have focused on the optimal code rate decision for joint source channel coding (JSCC). In
, a source coding distortion model and a channel-induced distortion model were proposed as a means of determining a combination of the optimal intra-refresh rate and code rate. Similarly, Frossard et al.
proposed an end-to-end distortion model, in which channel-induced distortion was assumed to be proportional to the number of lost pixels. Their modeling equations are very complex and include numerous modeling parameters. Kwon et al.
proposed a practical method based on an observation that the residual packet loss probability in the optimal code rate stays near to a constant value. However, actually this residual packet loss probability is not constant any more at the high channel packet loss probabilities.
In this paper, we propose an accurate estimation of channel loss threshold set for optimal code rate decision.
Ⅱ. Video Transmission System and Channel Model
The fundamental structure of the video transmissionsystem that supports both source and channel codings is presented in
. The system consists of a video encoder/decoder, a channel encoder/decoder, and rate allocation block. The rate allocation block optimizes the video quality for a given channel condition. The sender periodically receives information about the instantaneous channel status from the receiver through a RTCP feedback report. From this information, the block for the channel characteristic estimation calculates the current status of the channel packet loss rate,
, and the available total transmission rate,
. The channel encoder in
FEC packets for every
video packets by the RS code, denoted as RS(
). The code rate of RS code is defined as
비디오 전송 시스템의 기본 구조 Fig. 1. Fundamental structure of the video transmission system
be the source and channel coding bit rates. When we define the total transmission rate by
can be expressed as
) code, lost packets up to n-k packets in a transmission group can be completely recovered
. However, if more than
packets are lost, this transmission group cannot be recovered in its entirety. The residual packet loss probability is defined as the probability that a video packet is lost but cannot be correctly recovered even by the channel decoding. The residual packet loss probability denoted as
, and can be represented as
The overall distortion (
) of the reconstructed video depends upon both the source coding distortion (
) and channel induced distortion (
). We can also express the optimal code rate decision with respect to minimizing the overall distortion for the given
Ⅲ. Channel Loss Threshold Set
In this section, we introduce two types of loss threshold sets, the channel loss threshold set and the residual loss threshold set. We employed an H.264/AVC source coder and combined it with
=1,2,…,5) channel coder using R
=100 Kbps. Since we can adjust
can be redefined as follows:
, we can see that though
) barely decreases in the ranges of low
where almost all lost packets are recovered by the channel decoder. The optimal code rate can be classified as follows:
다양한 패킷손실 조건에서 Akiyo 시퀀스에 대한 최적의 부호율 결정을 위한 PSNR 곡선 Fig. 2. PSNR curves for determining optimum code rates in various packet loss channel condition for Akiyo sequence
We call the set of
values which correspond to the threshold boundary to change the code rate
a channel loss threshold set
). The CLTS is represented as
To get the CLTS quickly and accurately, we introduce another loss threshold set called residual loss threshold set(RLTS). We define the RLTS as
in which each element of RLTS,
, is the value shown in (3) for the given
Ⅳ. Joint Source-channel Distortion Model
If we use MSE as a distortion measure, we could observe that the difference of source distortion between
is likely equal to the channel distortion measured at
. We can formulate this observation as
) means the channel distortion.
, 0) can be represented as
) are the parameters of the model which depend on the encoded video sequence
shows that the distortion of the received video is linearly proportional to the residual packet loss probability. Therefore, by using (3), we can represent
n0=20일 때 k 값에 대한 잔여패킷 손실확률에 대비한 비디오 왜곡 Fig. 3. Video distortion versus residual packet loss probability PRkfor various k where n0=20
is a proportional constant. If we substitute (5), (10), and (11) into (9), we can obtain
where α and β are the model parameters. By using (12), we can get all elements of the RLTS defined in (8), and then the corresponding CLTS can be obtained from (3).
Ⅴ. Performance Evaluation
To evaluate the accuracy of the estimated CLTS using the joint source-channel distortion model, we performed simulations using H.264/AVC codec. The n
is fixed to be 20. We evaluate the PSNR performance of the code rates estimated by the proposed method, and compare it with the optimal code rate case when the packet loss ratio varies from 0.001 to 0.2.
compares the average PSNR performance. It can be seen that the PSNR performance obtained by the proposed method is very close to the optimal PSNR performance. The PSNR gap between the two approaches is just within 0.5 dB.
demonstrates the PSNR performance for various packet loss rates, in which the Soccer sequenceis used.
제안된 방법에 의해 추정된 부호율과 실제 최적의 부호율 간의 평균 PSNR 성능비교
Table 1. Comparison of the average PSNR between using real optimal code rates and using code rates estimated by the proposed method
Soccer 시퀀스에 대한 PSNR 성능 비교. Fig. 4. Comparison of PSNR performance for Soccer sequence
In this paper, we proposed an accurate estimation of channel loss threshold set for optimal FEC code rate decision.
“Joint source and channel coding for real time H.264 coded video transmission using mobile hand-held devices,”
IETE J. of Research
DOI : 10.4103/0377-2063.55984
“Analysis of video transmission over lossy channels,”
IEEE J. Sel. Areas in Commun.
DOI : 10.1109/49.848253
“Joint source/FEC rate selection for quality-optimal MPEG-2 video delivery,”
IEEE Trans. Image Process.
DOI : 10.1109/83.974566
“Adaptive code rate decision of joint source-channel coding for wireless video,”
DOI : 10.1049/el:20021163